Road accidents are, fortunately, rare events. Individual accidents are effectively unpredictable and occur generally because of road user error. Accident frequencies (at a site or by a driver) over a period of time are subject to considerable random variation but nevertheless display systematic features that can be analysed and, to some degree, explained. The talk starts with the obvious need for exposure measures in order to make meaningful and fair comparisons between accident numbers for different entities, whether sites, driver age groups, countries or time periods. It then moves on to look at the relationships between accidents, traffic flow and road design, as exemplified in the TRL series of junction accident studies. Whilst such predictive accident models are far from perfect, they do provide the potential for an intelligent assessment of how poorly a site is performing and hence of the possible need for remedial treatment. They also provide the means of carrying out, post hoc, an assessment of the effectiveness of a remedial treatment, by allowing for the ‘regression to mean’ effect in before and after comparisons, as applied in recent research on the effectiveness of speed cameras.

Want to know more about this project?